CD Projekt Red turns to AI voice tech for New Witcher 3 expansion, with one crucial caveat

More than a decade after The Witcher 3 originally debuted, CD Projekt Red is turning to artificial intelligence to solve a delicate casting dilemma for the game's upcoming expansion, Songs of the Past. Because Geralt’s Polish voice actor, Jacek Rozenek, experienced a life-altering stroke in 2019, the studio is utilising the AI vocal software Respeecher to bridge the gap between his current voice and his original 2014 performance.

AI is helping Geralt’s Polish voice actor return to The Witcher 3

As per Game File, Rozenek’s stroke temporarily robbed him of his speech, and he needed years of gruelling rehabilitation to recover. Although he managed to record a handful of lines for the 2022 next-gen update, the sessions left him deeply fatigued. Developers were determined to keep the iconic actor in the role for the new content, but faced a significant hurdle.

Rozenek’s vocal pitch has dropped considerably since the base game's production, creating a jarring mismatch for dialogue set within the same timeline.

To maintain continuity without overtaxing the actor, CD Projekt Red trained Respeecher’s AI model on Rozenek’s entire archive of original Witcher 3 recordings. Rather than applying a simple audio filter, the technology actively reshapes his new vocal takes, seamlessly restoring the cadence and timbre fans remember from 2014.

Respeecher synthetic speech artist Dmytro Vasylets told Game File that "getting there took dozens of hours, training models, going through the details with the CDPR team."

CDPR got Rozenek's blessing to use the tool, who told Game File, "I have great trust… for the creators of this game," and that, "when it comes to the dynamics of the voice, when it comes to the acting aspects, I feel very good, really very good."

CD Projekt Red was aware of the sensitive optics surrounding artificial intelligence and voice acting, and confirmed that this situation is strictly an exception. Localisation producer Mikołaj Szwed repoortedly confirmed that the studio would never use AI to modify a performer's voice without their explicit consent.

For more updates, join/follow our WhatsApp, Telegram and YouTube channels.